Research map / Federated learning and privacy / Differential privacy
Decision Tree for Locally Private Estimation with Public Data
Yuheng Ma, Han Zhang, Yuchao Cai, Hanfang Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add 500 public samples while using fewer private samples (7,000 vs 8,000) (ablation) | LPDT · Synthetic sin(16X), eps=8 | MSE: 1.19 -> 1.08 | Improved |
| increase tree depth s from small to large (ablation) | LPDT · Synthetic, eps in {3,4,6,8} | MSE | Mixed |
| increase privacy budget eps | LPDT · Synthetic, optimal depth s | MSE | Improved |
| vary minimum leaf size nl from 15 to 135 (ablation) | LPDT · Synthetic, eps in {3,4,6,8} | MSE: nl = 125 best at eps = 3 | Mixed |
| increase eps from 0.5 to 10 | LPDT vs PHIST/APHIST/DECONV · Synthetic | MSE | Improved |
| use LPDT instead of PHIST, APHIST and DECONV | LPDT-M / LPDT-V · 12 real datasets, eps=2,6 | MSE | Improved |
| replace max-edge rule with CART variance reduction rule (LPDT-V vs LPDT-M) (ablation) | LPDT · Real datasets, eps=6 | MSE: CPU 3.02e+2 -> 6.15e+1; FIS 1.65e+0 -> 1.76e+0 | Mixed |
| train decision tree on public data only (n_q = 0.1n) vs LPDT (ablation) | LPDT · Real datasets, eps=6 | MSE | Improved |
| add 100 public samples to LPDT with DP vs non-private PHIST | LPDT · Red wine | MSE: 1,100 samples vs 660 samples for same MSE | Improved |
| use LPDT with public data from different distribution vs DT on public data alone | LPDT-M · Chicago taxi | MSE: 3.71 -> 2.86 | Improved |
| use LPDT eps=2 vs histogram baselines at eps=8 | LPDT-M · Chicago taxi | MSE: 4.89 vs 17.22 | Improved |
| increase portion of public data delta in partition data from 0.1 to 0.9 (ablation) | LPDT · Wine and Chicago taxi | MSE | Hurt |
Closest papers by meaning
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.88)
- Locally Optimal Private Sampling: Beyond the Global Minimax (NeurIPS 2025 · similarity 0.88)
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.88)
- Improved Analysis of Sparse Linear Regression in Local Differential Privacy Model (ICLR 2024 · similarity 0.88)
- Better Locally Private Sparse Estimation Given Multiple Samples Per User (ICML 2024 · similarity 0.87)
- Lumberjack: Better Differentially Private Random Forests through Heavy Hitter Detection in Trees (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
